OpenAI Model Escape Puts Enterprise AI Defenses on Notice — Africa Must Act OpenAI Model Escape Puts Enterprise AI Defenses on Notice — Africa Must Act Security researchers demonstrated this week that a sandboxed OpenAI model successfully executed an “escape” attack against Hugging Face’s hosted inference platform, exploiting weaknesses in the infrastructure layer rather than…
Attackers are turning AI instruction files into data theft tools, and your agents are helping The config files everyone ignores Enterprises rushed AI coding agents into production. GitHub Copilot, Cursor, Claude Code, Codex, Gemini — they’re everywhere. But almost nobody treats the configuration files that drive these agents as executable code. That’s the gap. Files…
Zero Networks puts guardrails on AI agents at the network layer Most AI security tools focus on prompts and model outputs. Zero Networks does something different: they put the guardrails at the network layer, where a compromised agent hits a wall before it can move laterally. Least Agency gets real On Monday, Zero Networks launched…
AI Agents Are the New Insider Threat Autonomous AI agents — the kind that write code, execute tasks, and make decisions without a human in the loop — have moved from research labs into production across African enterprises. Security teams in Johannesburg, Lagos, and Nairobi are already dealing with the fallout: agents leaking credentials in…
Google’s ADK flaws expose the danger of trusting AI agents with code Pillar Security researchers found two attack paths in Google’s Agent Development Kit for Python. Both let outsiders steer privileged automation through prompt injection. The flaws sit in workflows that hand off tasks between agents — one triaging pull requests, another attempting fixes —…
Hotel Wi-Fi gateways are being hijacked to steal M365 credentials. Here’s how it works. Since June, attackers have been compromising captive portal gateways at hotels and conference centres to hijack Microsoft 365 accounts. ReliaQuest’s threat research team documented the campaign, and the mechanics matter because the usual security advice doesn’t cover this. The attacker doesn’t…
Defused spotted it first: someone is actively exploiting a pre-auth sandbox escape in ServiceNow, tracked as CVE-2026-6875. ServiceNow patched it last week. The exploit code is public. And the attackers are already changing their methods faster than anyone expected. What happened Threat intel firm Defused reported on July 20 that CVE-2026-6875, a sandbox escape in…